Verified on Two Vendors: When the Verdict Flips on Physics Alone
Three posts in, we've shown a verifier that recomputes Bell/Mermin statistics client-side (post 1), works across witness families with zero code changes (post 2), and honestly rejects real quantum runs whose hardware fidelity falls below the physics threshold (post 3).
The Cepheus Svetlichny REJECTs from the last post raised the natural question: is Svetlichny impossible to verify on our stack, or was that just Cepheus?
We ran the same experiment on IonQ Forte-1.
Same code. Same operator. Same block-binding protocol. Different QPU.
Point our submission code at arn:aws:braket:us-east-1::device/qpu/ionq/Forte-1 instead of arn:aws:braket:us-west-1::device/qpu/rigetti/Cepheus-1-108Q. Change nothing else. Every other field in the anchor — the Svetlichny operator, the tilted measurement angles, the challenge derivation, the KDF-derived per-shot settings, the Merkle-root construction, the entry_hash preimage — is byte-identical to the Cepheus attempts.
The result: |Sv| = 35439195501211/6604444890780 ≈ 5.366. VERIFIED.
Above the classical bound of 4 by 34%. Above the biseparability bound (the strong claim: genuine tripartite entanglement). 94.9% of the way to the Tsirelson ceiling of 4√2 ≈ 5.657.
The controlled experiment
Three of the Svetlichny anchors on qblex.com/verify form a controlled scientific experiment:
| QPU | Sv | Verdict | Hardware fidelity |
|---|---|---|---|
| Rigetti Cepheus | 3.310 | REJECT | 58.5% |
| Rigetti Cepheus | 3.469 | REJECT | 61.3% |
| IonQ Forte-1 | 5.366 | PASS | 94.9% |
Same Svetlichny operator across all three. Same tilted measurement angles. Same challenge-binding to the anchor's block-derived nonce. Same anchor format. Same verifier code path.
The only variable that changes between REJECT and PASS is the device.
That's what "the verifier is a physics discriminant" means in the most concrete possible sense. The pass/reject verdict flips because the underlying quantum hardware has different 3-qubit fidelity. Not because our code chose a friendlier witness. Not because we tuned a threshold. Not because we cherry-picked shots. The verdict tracks physics.
The fidelity budget
Every Bell-family inequality has the same fidelity threshold to violate: 1/√2 ≈ 70.7% (see the last post for the derivation). Backed out from the measured statistics on our page:
| QPU and circuit | Effective fidelity | Above 70.7%? |
|---|---|---|
| Cepheus, 2-qubit CHSH | ~75% | Yes, passes CHSH |
| Cepheus, 3-qubit Mermin | ~65-68% | Just below; Mermin passes on looser statistics |
| Cepheus, 3-qubit Svetlichny tilted | 58-61% | No, rejects |
| IonQ Forte-1, 3-qubit Svetlichny tilted | 94.9% | Yes, passes decisively |
Cepheus's 3-qubit fidelity straddles the threshold. IonQ Forte-1's has 24 percentage points of margin.
Neither vendor is "better" in an absolute sense — each has strengths, cost profiles, and problems it fits. But for demonstrating genuine tripartite entanglement via Svetlichny violation, IonQ Forte-1 is where the physics currently lives. And our verifier proves it, with an anchor a reader can click and reproduce byte-for-byte in their browser.
Why this matters
One: the verifier scales to any quantum hardware. Rigetti superconducting today, IonQ trapped-ion today, whatever comes next — same anchor format, same client-side verification, same cryptographic guarantees. As QPU fidelities cross more thresholds, the same code path just starts passing more inequalities.
Two: the anchors are witness-agnostic and vendor-agnostic. The verify page has nine anchors across two vendors and three witness families right now. Adding a tenth from a third vendor is a fifteen-line diff.
Three: the failure mode is legible. When a run REJECTs, you can compute exactly why: what fidelity would have cleared the threshold, what circuit depth introduced the noise, what physics is in the way. That's a very different failure mode from "the black box said no."
Try it
Go to qblex.com/verify. Click the Svetlichny buttons — the two Cepheus REJECTs, then the IonQ PASS at 5.366. Watch the verifier apply the same computation to all three and reach two different verdicts. Nothing in the code path changes. The physics does.
Nine anchors on the page. Two QPU vendors. Three witness families. Five PASS, four REJECT. Every one a real hardware run, cryptographically committed, byte-for-byte verified externally, redacted of any account identifier, and reproducible in your browser without contacting our servers.
Don't trust us. Run the math.
QBLEX technologies are patent pending (U.S. App. Nos. 19/711,486; 19/702,980; 19/707,649 — priority June 2025). "Patent pending" does not indicate a granted patent. Certain aspects of QBLEX technology are protected as trade secrets and are not disclosed in any public filing.